File size: 13,328 Bytes
881f9f2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 | """Upstream AssetOpsBench OpenAIAgentRunner smoke wrapper.
This is intentionally separate from ``scripts/aat_runner.py``. The production
Experiment 1 Cell A/B path uses the team-local OpenAI Agents SDK wrapper so the
direct and MCP arms share the same agent loop. This script is only for the
parity smoke required by #104: instantiate AssetOpsBench's upstream
``OpenAIAgentRunner`` Python API with this repo's Smart Grid MCP server paths,
then translate the result into the benchmark artifact schema.
The upstream ``openai-agent`` CLI does not expose a server-path override, so the
Python API is the narrowest way to run the upstream runner against Smart Grid.
The Python API still assumes AOB's own MCP server entry points; for this smoke
we patch its MCP server factory to launch this repo's Smart Grid servers through
the warmed Insomnia Python/bootstrap path used by the benchmark runner.
"""
from __future__ import annotations
import argparse
import asyncio
import importlib.util
import json
import logging
import os
import shlex
import sys
import time
import types
from pathlib import Path
from typing import Any
_REPO_ROOT = Path(__file__).resolve().parent.parent
if str(_REPO_ROOT) not in sys.path:
sys.path.insert(0, str(_REPO_ROOT))
_LOG = logging.getLogger("aat_upstream_openai_runner")
SERVER_PATHS = {
"iot": "mcp_servers/iot_server/server.py",
"fmsr": "mcp_servers/fmsr_server/server.py",
"tsfm": "mcp_servers/tsfm_server/server.py",
"wo": "mcp_servers/wo_server/server.py",
}
def build_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(
prog="aat_upstream_openai_runner",
description="Run AOB's upstream OpenAIAgentRunner against Smart Grid MCP servers.",
)
parser.add_argument("--prompt", required=True)
parser.add_argument("--output", required=True)
parser.add_argument("--model-id", required=True)
parser.add_argument("--aob-path", required=True)
parser.add_argument("--max-turns", type=int, default=30)
parser.add_argument("--verbose", action="store_true")
return parser
def _setup_logging(verbose: bool) -> None:
logging.basicConfig(
level=logging.DEBUG if verbose else logging.INFO,
format="%(asctime)s %(levelname)s %(name)s: %(message)s",
)
def _bootstrap_aob(aob_path: Path) -> None:
src = aob_path / "src"
if not src.exists():
raise FileNotFoundError(f"AssetOpsBench src/ not found under {aob_path}")
if str(src) not in sys.path:
sys.path.insert(0, str(src))
_install_unused_claude_sdk_stub()
def _install_unused_claude_sdk_stub() -> None:
"""Avoid AOB's top-level Claude runner import for this OpenAI-only smoke.
``agent.openai_agent.runner`` is the only upstream surface used here, but
importing it normally first executes ``agent.__init__``, which imports the
Claude runner and requires ``claude_agent_sdk``. That SDK is irrelevant to
the OpenAI parity smoke and is not available in the Insomnia AaT runtime.
"""
if "claude_agent_sdk" in sys.modules:
return
if importlib.util.find_spec("claude_agent_sdk") is not None:
_LOG.warning(
"real claude_agent_sdk package is installed; not installing the "
"OpenAI parity smoke import shim"
)
return
stub = types.ModuleType("claude_agent_sdk")
for name in (
"AssistantMessage",
"ClaudeAgentOptions",
"HookMatcher",
"ResultMessage",
"TextBlock",
"ToolUseBlock",
):
setattr(stub, name, type(name, (), {}))
async def _query(*_args: object, **_kwargs: object) -> None:
raise RuntimeError(
"claude_agent_sdk stub is unavailable in OpenAI parity smoke"
)
stub.query = _query
sys.modules["claude_agent_sdk"] = stub
def _smartgrid_server_paths(repo_root: Path) -> dict[str, Path]:
paths: dict[str, Path] = {}
for name, rel in SERVER_PATHS.items():
path = repo_root / rel
if not path.exists():
raise FileNotFoundError(f"Smart Grid MCP server missing: {path}")
paths[name] = path
return paths
def _parse_parallel_tool_calls() -> bool | None:
raw = os.environ.get("AAT_PARALLEL_TOOL_CALLS", "false").strip().lower()
if raw in {"", "false", "0", "no", "off"}:
return False
if raw in {"true", "1", "yes", "on"}:
return True
if raw in {"auto", "default", "none"}:
return None
raise ValueError(
"AAT_PARALLEL_TOOL_CALLS must be true, false, or auto; " f"got {raw!r}"
)
def _is_watsonx_model(model_id: str) -> bool:
return model_id.strip().lower().startswith("watsonx/")
def _configure_litellm_provider_compat(model_id: str) -> None:
if not _is_watsonx_model(model_id):
return
try:
import litellm # type: ignore
except ImportError:
return
litellm.drop_params = True
def _patch_aob_openai_runner(
aob_openai_runner: Any,
repo_root: Path,
model_id: str = "",
) -> list[str]:
"""Patch AOB runner dependencies while leaving OpenAIAgentRunner.run intact."""
from scripts.aat_system_prompt import AOB_SOURCE_SHA
patches: list[str] = []
build_mcp_servers = getattr(aob_openai_runner, "_build_mcp_servers", None)
if not callable(build_mcp_servers):
raise RuntimeError(
"AssetOpsBench OpenAIAgentRunner patch precondition failed: "
"agent.openai_agent.runner._build_mcp_servers is missing or not "
f"callable at expected AOB source SHA {AOB_SOURCE_SHA}. "
"Refusing to run parity smoke with the default upstream MCP launcher."
)
if not callable(getattr(aob_openai_runner, "Agent", None)):
raise RuntimeError(
"AssetOpsBench OpenAIAgentRunner patch precondition failed: "
"agent.openai_agent.runner.Agent is missing or not callable at "
f"expected AOB source SHA {AOB_SOURCE_SHA}. Refusing to run parity "
"smoke without the local-vLLM parallel_tool_calls setting."
)
from agents import Agent as SDKAgent, ModelSettings
from agents.mcp import MCPServerStdio
from scripts.aat_tools_mcp import _client_timeout_seconds, _server_params
def _build_smartgrid_mcp_servers(
server_paths: dict[str, Path | str],
) -> list[MCPServerStdio]:
client_timeout = _client_timeout_seconds()
servers: list[MCPServerStdio] = []
for name, spec in server_paths.items():
path = Path(spec)
if not path.is_absolute():
path = repo_root / path
params = _server_params(repo_root, path)
command_line = [
str(params["command"]),
*[str(arg) for arg in params["args"]],
]
print(
"Upstream parity MCP server "
f"{name}: timeout={client_timeout:g}s {shlex.join(command_line)}",
file=sys.stderr,
)
servers.append(
MCPServerStdio(
name=name,
params=params,
cache_tools_list=True,
client_session_timeout_seconds=client_timeout,
)
)
return servers
parallel_tool_calls = _parse_parallel_tool_calls()
if _is_watsonx_model(model_id) and parallel_tool_calls is True:
raise ValueError(
"WatsonX does not support parallel_tool_calls; set "
"AAT_PARALLEL_TOOL_CALLS=false or auto for hosted WatsonX runs."
)
_configure_litellm_provider_compat(model_id)
effective_parallel_tool_calls = parallel_tool_calls
if _is_watsonx_model(model_id) and effective_parallel_tool_calls is False:
effective_parallel_tool_calls = None
def _agent_with_model_settings(*args: Any, **kwargs: Any) -> Any:
if effective_parallel_tool_calls is not None:
kwargs.setdefault(
"model_settings",
ModelSettings(parallel_tool_calls=parallel_tool_calls),
)
return SDKAgent(*args, **kwargs)
aob_openai_runner._build_mcp_servers = _build_smartgrid_mcp_servers
patches.append("mcp_server_launch")
aob_openai_runner.Agent = _agent_with_model_settings
patches.append(f"parallel_tool_calls={parallel_tool_calls}")
if _is_watsonx_model(model_id):
patches.append("watsonx_drop_unsupported_params")
return patches
def _tool_call_payload(tool_call: Any) -> dict[str, Any]:
return {
"name": getattr(tool_call, "name", "") or "",
"arguments": getattr(tool_call, "input", {}) or {},
"call_id": getattr(tool_call, "id", "") or "",
"output": getattr(tool_call, "output", None),
}
def _serialize_result(
*,
args: argparse.Namespace,
prompt: str,
result: Any,
duration_seconds: float,
server_paths: dict[str, Path],
patches: list[str],
) -> dict[str, Any]:
trajectory = getattr(result, "trajectory", None)
turns = list(getattr(trajectory, "turns", []) or [])
history = []
tool_call_count = 0
for index, turn in enumerate(turns, start=1):
tool_calls = [
_tool_call_payload(call)
for call in list(getattr(turn, "tool_calls", []) or [])
]
tool_call_count += len(tool_calls)
history.append(
{
"turn": index,
"role": "assistant",
"content": getattr(turn, "text", "") or "",
"tool_calls": tool_calls,
"input_tokens": getattr(turn, "input_tokens", 0) or 0,
"output_tokens": getattr(turn, "output_tokens", 0) or 0,
}
)
answer = getattr(result, "answer", "") or ""
max_turns_reached = bool(
getattr(result, "max_turns_reached", False)
or getattr(result, "max_turns_exhausted", False)
)
return {
"question": prompt,
"answer": answer,
"success": (not max_turns_reached) and bool(answer),
"failed_tools": [],
"max_turns_exhausted": max_turns_reached,
"turn_count": len(history),
"tool_call_count": tool_call_count,
"history": history,
"runner_meta": {
"runner": "AssetOpsBench OpenAIAgentRunner",
"runner_source": "upstream_python_api",
"aob_path": str(Path(args.aob_path).resolve()),
"model_id": args.model_id,
"max_turns": args.max_turns,
"server_paths": {name: str(path) for name, path in server_paths.items()},
"import_shims": ["claude_agent_sdk"],
"aob_runner_patches": patches,
"duration_seconds": duration_seconds,
},
}
def _write_output(path: Path, payload: dict[str, Any]) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(json.dumps(payload, indent=2) + "\n", encoding="utf-8")
async def _main(args: argparse.Namespace) -> int:
repo_root = Path(__file__).resolve().parent.parent
aob_path = Path(args.aob_path).resolve()
server_paths = _smartgrid_server_paths(repo_root)
_bootstrap_aob(aob_path)
from agent.openai_agent import runner as aob_openai_runner
patches = _patch_aob_openai_runner(aob_openai_runner, repo_root, args.model_id)
OpenAIAgentRunner = aob_openai_runner.OpenAIAgentRunner
runner = OpenAIAgentRunner(
server_paths=server_paths,
model=args.model_id,
max_turns=args.max_turns,
)
start = time.time()
try:
result = await runner.run(args.prompt)
except Exception as exc:
_LOG.exception("upstream OpenAIAgentRunner failed: %s", exc)
_write_output(
Path(args.output),
{
"question": args.prompt,
"answer": "",
"success": False,
"error": f"{type(exc).__name__}: {exc}",
"failed_tools": [],
"max_turns_exhausted": False,
"turn_count": 0,
"tool_call_count": 0,
"history": [],
"runner_meta": {
"runner": "AssetOpsBench OpenAIAgentRunner",
"runner_source": "upstream_python_api",
"aob_path": str(aob_path),
"model_id": args.model_id,
"max_turns": args.max_turns,
"server_paths": {
name: str(path) for name, path in server_paths.items()
},
"import_shims": ["claude_agent_sdk"],
"aob_runner_patches": patches,
"duration_seconds": time.time() - start,
},
},
)
return 1
payload = _serialize_result(
args=args,
prompt=args.prompt,
result=result,
duration_seconds=time.time() - start,
server_paths=server_paths,
patches=patches,
)
_write_output(Path(args.output), payload)
return 0 if payload["success"] else 1
def main() -> None:
args = build_parser().parse_args()
_setup_logging(args.verbose)
sys.exit(asyncio.run(_main(args)))
if __name__ == "__main__":
main()
|